Conformational analysis of protonated N-acetyl hexosamines: unexpected methylation effects from first-principles and machine learning

Abstract

Elucidating the intricate 3D conformational behavior of inherently flexible carbohydrates is crucial for understanding their biological functions, yet it remains experimentally challenging. While traditional ab initio computational approaches, such as density functional theory (DFT), can sample low-energy conformers, they are often resource-intensive. In this work, we developed and employed machine learning-driven methods that efficiently locate low-energy candidate structures by leveraging previously established local minima databases. These candidates were then reoptimized using a target ab initio method, specifically DFT, by training neural network potential (NNP) models to mimic the DFT potential energy surface. We successfully applied this approach to elucidate the 3D structures of protonated N-acetyl hexosamines (HexNAcH⁺) and their methylated forms, resulting in a comprehensive structural database of 32 monosaccharides with first-principles accuracy. Although our findings generally align with existing literature, the results revealed unexpected methylation effects that challenge the current understanding of HexNAcH⁺ conformational behavior. More importantly, based on the experimental vibrational spectra obtained via infrared multiple photon dissociation (IRMPD) from literature (for GlcNAcH⁺, GalNAcH⁺, and ManNAcH⁺) and our simulated spectra of all 16 HexNAcH⁺ structures, we find reasonable expectation that the remaining experimentally unexplored HexNAcH⁺ can be resolved via IRMPD.